Description

PyTensor is missing this simple rewrite:

import pytensor.tensor as pt
from pytensor.graph import rewrite_graph

x = pt.scalar("x")
out = pt.softplus(pt.log(x))
new_out = rewrite_graph(out, include=("canonicalize", "stabilize", "specialize"))
new_out.dprint()

I am ready to propose a PR that addresses this issue.

I could also extend it to cover the case softplus(-log(x)) -> log1p(1/x). However I have noticed that even the simple exp(-log(x)) -> 1/x is missing, so I wonder if there is an underlying reason to avoid such simplifications, that I am not aware of.